Linaro-mm-sig September 2021

linaro-mm-sig@lists.linaro.org

12 participants
35 discussions

by Christian König

Hi guys, we are currently working an Freesync and direct scan out from system memory on AMD APUs in A+A laptops. On problem we stumbled over is that our display hardware needs to scan out from uncached system memory and we currently don't have a way to communicate that through DMA-buf. For our specific use case at hand we are going to implement something driver specific, but the question is should we have something more generic for this? After all the system memory access pattern is a PCIe extension and as such something generic. Regards, Christian.

1 year, 8 months

[PATCH] dma-buf/poll: Get a file reference for outstanding fence callbacks

by Michel Dänzer

From: Michel Dänzer <mdaenzer(a)redhat.com> This makes sure we don't hit the BUG_ON(dmabuf->cb_in.active || dmabuf->cb_out.active); in dma_buf_release, which could be triggered by user space closing the dma-buf file description while there are outstanding fence callbacks from dma_buf_poll. Cc: stable(a)vger.kernel.org Signed-off-by: Michel Dänzer <mdaenzer(a)redhat.com> --- drivers/dma-buf/dma-buf.c | 18 ++++++++++++------ 1 file changed, 12 insertions(+), 6 deletions(-) diff --git a/drivers/dma-buf/dma-buf.c b/drivers/dma-buf/dma-buf.c index 6c520c9bd93c..ec25498a971f 100644 --- a/drivers/dma-buf/dma-buf.c +++ b/drivers/dma-buf/dma-buf.c @@ -65,12 +65,9 @@ static void dma_buf_release(struct dentry *dentry) BUG_ON(dmabuf->vmapping_counter); /* - * Any fences that a dma-buf poll can wait on should be signaled - * before releasing dma-buf. This is the responsibility of each - * driver that uses the reservation objects. - * - * If you hit this BUG() it means someone dropped their ref to the - * dma-buf while still having pending operation to the buffer. + * If you hit this BUG() it could mean: + * * There's a file reference imbalance in dma_buf_poll / dma_buf_poll_cb or somewhere else + * * dmabuf->cb_in/out.active are non-0 despite no pending fence callback */ BUG_ON(dmabuf->cb_in.active || dmabuf->cb_out.active); @@ -196,6 +193,7 @@ static loff_t dma_buf_llseek(struct file *file, loff_t offset, int whence) static void dma_buf_poll_cb(struct dma_fence *fence, struct dma_fence_cb *cb) { struct dma_buf_poll_cb_t *dcb = (struct dma_buf_poll_cb_t *)cb; + struct dma_buf *dmabuf = container_of(dcb->poll, struct dma_buf, poll); unsigned long flags; spin_lock_irqsave(&dcb->poll->lock, flags); @@ -203,6 +201,8 @@ static void dma_buf_poll_cb(struct dma_fence *fence, struct dma_fence_cb *cb) dcb->active = 0; spin_unlock_irqrestore(&dcb->poll->lock, flags); dma_fence_put(fence); + /* Paired with get_file in dma_buf_poll */ + fput(dmabuf->file); } static bool dma_buf_poll_shared(struct dma_resv *resv, @@ -278,6 +278,9 @@ static __poll_t dma_buf_poll(struct file *file, poll_table *poll) spin_unlock_irq(&dmabuf->poll.lock); if (events & EPOLLOUT) { + /* Paired with fput in dma_buf_poll_cb */ + get_file(dmabuf->file); + if (!dma_buf_poll_shared(resv, dcb) && !dma_buf_poll_excl(resv, dcb)) /* No callback queued, wake up any other waiters */ @@ -299,6 +302,9 @@ static __poll_t dma_buf_poll(struct file *file, poll_table *poll) spin_unlock_irq(&dmabuf->poll.lock); if (events & EPOLLIN) { + /* Paired with fput in dma_buf_poll_cb */ + get_file(dmabuf->file); + if (!dma_buf_poll_excl(resv, dcb)) /* No callback queued, wake up any other waiters */ dma_buf_poll_cb(NULL, &dcb->cb); -- 2.32.0

2 years, 5 months

Re: [Linaro-mm-sig] [RFC PATCH v2 2/2] RDMA/rxe: Add dma-buf support

by Daniel Vetter

On Wed, Sep 29, 2021 at 01:19:05PM +0900, Shunsuke Mie wrote: > Implement a ib device operation ‘reg_user_mr_dmabuf’. Generate a > rxe_map from the memory space linked the passed dma-buf. > > Signed-off-by: Shunsuke Mie <mie(a)igel.co.jp> > --- > drivers/infiniband/sw/rxe/rxe_loc.h | 2 + > drivers/infiniband/sw/rxe/rxe_mr.c | 118 ++++++++++++++++++++++++++ > drivers/infiniband/sw/rxe/rxe_verbs.c | 34 ++++++++ > drivers/infiniband/sw/rxe/rxe_verbs.h | 2 + > 4 files changed, 156 insertions(+) > > diff --git a/drivers/infiniband/sw/rxe/rxe_loc.h b/drivers/infiniband/sw/rxe/rxe_loc.h > index 1ca43b859d80..8bc19ea1a376 100644 > --- a/drivers/infiniband/sw/rxe/rxe_loc.h > +++ b/drivers/infiniband/sw/rxe/rxe_loc.h > @@ -75,6 +75,8 @@ u8 rxe_get_next_key(u32 last_key); > void rxe_mr_init_dma(struct rxe_pd *pd, int access, struct rxe_mr *mr); > int rxe_mr_init_user(struct rxe_pd *pd, u64 start, u64 length, u64 iova, > int access, struct rxe_mr *mr); > +int rxe_mr_dmabuf_init_user(struct rxe_pd *pd, int fd, u64 start, u64 length, > + u64 iova, int access, struct rxe_mr *mr); > int rxe_mr_init_fast(struct rxe_pd *pd, int max_pages, struct rxe_mr *mr); > int rxe_mr_copy(struct rxe_mr *mr, u64 iova, void *addr, int length, > enum rxe_mr_copy_dir dir); > diff --git a/drivers/infiniband/sw/rxe/rxe_mr.c b/drivers/infiniband/sw/rxe/rxe_mr.c > index 53271df10e47..af6ef671c3a5 100644 > --- a/drivers/infiniband/sw/rxe/rxe_mr.c > +++ b/drivers/infiniband/sw/rxe/rxe_mr.c > @@ -4,6 +4,7 @@ > * Copyright (c) 2015 System Fabric Works, Inc. All rights reserved. > */ > > +#include <linux/dma-buf.h> > #include "rxe.h" > #include "rxe_loc.h" > > @@ -245,6 +246,120 @@ int rxe_mr_init_user(struct rxe_pd *pd, u64 start, u64 length, u64 iova, > return err; > } > > +static int rxe_map_dmabuf_mr(struct rxe_mr *mr, > + struct ib_umem_dmabuf *umem_dmabuf) > +{ > + struct rxe_map_set *set; > + struct rxe_phys_buf *buf = NULL; > + struct rxe_map **map; > + void *vaddr, *vaddr_end; > + int num_buf = 0; > + int err; > + size_t remain; > + > + mr->dmabuf_map = kzalloc(sizeof &mr->dmabuf_map, GFP_KERNEL); dmabuf_maps are just tagged pointers (and we could shrink them to actually just a tagged pointer if anyone cares about the overhead of the separate bool), allocating them seperately is overkill. > + if (!mr->dmabuf_map) { > + err = -ENOMEM; > + goto err_out; > + } > + > + err = dma_buf_vmap(umem_dmabuf->dmabuf, mr->dmabuf_map); > + if (err) > + goto err_free_dmabuf_map; > + > + set = mr->cur_map_set; > + set->page_shift = PAGE_SHIFT; > + set->page_mask = PAGE_SIZE - 1; > + > + map = set->map; > + buf = map[0]->buf; > + > + vaddr = mr->dmabuf_map->vaddr; dma_buf_map can be an __iomem too, you shouldn't dig around in this, but use the dma-buf-map.h helpers instead. On x86 (and I think also on most arm) it doesn't matter, but it's kinda not very nice in a pure software driver. If anything is missing in dma-buf-map.h wrappers just add more. Or alternatively you need to fail the import if you can't handle __iomem. Aside from these I think the dma-buf side here for cpu access looks reasonable now. -Daniel > + vaddr_end = vaddr + umem_dmabuf->dmabuf->size; > + remain = umem_dmabuf->dmabuf->size; > + > + for (; remain; vaddr += PAGE_SIZE) { > + if (num_buf >= RXE_BUF_PER_MAP) { > + map++; > + buf = map[0]->buf; > + num_buf = 0; > + } > + > + buf->addr = (uintptr_t)vaddr; > + if (remain >= PAGE_SIZE) > + buf->size = PAGE_SIZE; > + else > + buf->size = remain; > + remain -= buf->size; > + > + num_buf++; > + buf++; > + } > + > + return 0; > + > +err_free_dmabuf_map: > + kfree(mr->dmabuf_map); > +err_out: > + return err; > +} > + > +static void rxe_unmap_dmabuf_mr(struct rxe_mr *mr) > +{ > + struct ib_umem_dmabuf *umem_dmabuf = to_ib_umem_dmabuf(mr->umem); > + > + dma_buf_vunmap(umem_dmabuf->dmabuf, mr->dmabuf_map); > + kfree(mr->dmabuf_map); > +} > + > +int rxe_mr_dmabuf_init_user(struct rxe_pd *pd, int fd, u64 start, u64 length, > + u64 iova, int access, struct rxe_mr *mr) > +{ > + struct ib_umem_dmabuf *umem_dmabuf; > + struct rxe_map_set *set; > + int err; > + > + umem_dmabuf = ib_umem_dmabuf_get(pd->ibpd.device, start, length, fd, > + access, NULL); > + if (IS_ERR(umem_dmabuf)) { > + err = PTR_ERR(umem_dmabuf); > + goto err_out; > + } > + > + rxe_mr_init(access, mr); > + > + err = rxe_mr_alloc(mr, ib_umem_num_pages(&umem_dmabuf->umem), 0); > + if (err) { > + pr_warn("%s: Unable to allocate memory for map\n", __func__); > + goto err_release_umem; > + } > + > + mr->ibmr.pd = &pd->ibpd; > + mr->umem = &umem_dmabuf->umem; > + mr->access = access; > + mr->state = RXE_MR_STATE_VALID; > + mr->type = IB_MR_TYPE_USER; > + > + set = mr->cur_map_set; > + set->length = length; > + set->iova = iova; > + set->va = start; > + set->offset = ib_umem_offset(mr->umem); > + > + err = rxe_map_dmabuf_mr(mr, umem_dmabuf); > + if (err) > + goto err_free_map_set; > + > + return 0; > + > +err_free_map_set: > + rxe_mr_free_map_set(mr->num_map, mr->cur_map_set); > +err_release_umem: > + ib_umem_release(&umem_dmabuf->umem); > +err_out: > + return err; > +} > + > int rxe_mr_init_fast(struct rxe_pd *pd, int max_pages, struct rxe_mr *mr) > { > int err; > @@ -703,6 +818,9 @@ void rxe_mr_cleanup(struct rxe_pool_entry *arg) > { > struct rxe_mr *mr = container_of(arg, typeof(*mr), pelem); > > + if (mr->umem && mr->umem->is_dmabuf) > + rxe_unmap_dmabuf_mr(mr); > + > ib_umem_release(mr->umem); > > if (mr->cur_map_set) > diff --git a/drivers/infiniband/sw/rxe/rxe_verbs.c b/drivers/infiniband/sw/rxe/rxe_verbs.c > index 9d0bb9aa7514..6191bb4f434d 100644 > --- a/drivers/infiniband/sw/rxe/rxe_verbs.c > +++ b/drivers/infiniband/sw/rxe/rxe_verbs.c > @@ -916,6 +916,39 @@ static struct ib_mr *rxe_reg_user_mr(struct ib_pd *ibpd, > return ERR_PTR(err); > } > > +static struct ib_mr *rxe_reg_user_mr_dmabuf(struct ib_pd *ibpd, u64 start, > + u64 length, u64 iova, int fd, > + int access, struct ib_udata *udata) > +{ > + int err; > + struct rxe_dev *rxe = to_rdev(ibpd->device); > + struct rxe_pd *pd = to_rpd(ibpd); > + struct rxe_mr *mr; > + > + mr = rxe_alloc(&rxe->mr_pool); > + if (!mr) { > + err = -ENOMEM; > + goto err2; > + } > + > + rxe_add_index(mr); > + > + rxe_add_ref(pd); > + > + err = rxe_mr_dmabuf_init_user(pd, fd, start, length, iova, access, mr); > + if (err) > + goto err3; > + > + return &mr->ibmr; > + > +err3: > + rxe_drop_ref(pd); > + rxe_drop_index(mr); > + rxe_drop_ref(mr); > +err2: > + return ERR_PTR(err); > +} > + > static struct ib_mr *rxe_alloc_mr(struct ib_pd *ibpd, enum ib_mr_type mr_type, > u32 max_num_sg) > { > @@ -1081,6 +1114,7 @@ static const struct ib_device_ops rxe_dev_ops = { > .query_qp = rxe_query_qp, > .query_srq = rxe_query_srq, > .reg_user_mr = rxe_reg_user_mr, > + .reg_user_mr_dmabuf = rxe_reg_user_mr_dmabuf, > .req_notify_cq = rxe_req_notify_cq, > .resize_cq = rxe_resize_cq, > > diff --git a/drivers/infiniband/sw/rxe/rxe_verbs.h b/drivers/infiniband/sw/rxe/rxe_verbs.h > index c807639435eb..0aa95ab06b6e 100644 > --- a/drivers/infiniband/sw/rxe/rxe_verbs.h > +++ b/drivers/infiniband/sw/rxe/rxe_verbs.h > @@ -334,6 +334,8 @@ struct rxe_mr { > > struct rxe_map_set *cur_map_set; > struct rxe_map_set *next_map_set; > + > + struct dma_buf_map *dmabuf_map; > }; > > enum rxe_mw_state { > -- > 2.17.1 > -- Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch

2 years, 6 months

Re: [Linaro-mm-sig] [PATCH v6 0/2] Add p2p via dmabuf to habanalabs

by Daniel Vetter

On Tue, Sep 28, 2021 at 10:04:29AM +0300, Oded Gabbay wrote: > On Thu, Sep 23, 2021 at 12:22 PM Oded Gabbay <ogabbay(a)kernel.org> wrote: > > > > On Sat, Sep 18, 2021 at 11:38 AM Oded Gabbay <ogabbay(a)kernel.org> wrote: > > > > > > On Fri, Sep 17, 2021 at 3:30 PM Daniel Vetter <daniel(a)ffwll.ch> wrote: > > > > > > > > On Thu, Sep 16, 2021 at 10:10:14AM -0300, Jason Gunthorpe wrote: > > > > > On Thu, Sep 16, 2021 at 02:31:34PM +0200, Daniel Vetter wrote: > > > > > > On Wed, Sep 15, 2021 at 10:45:36AM +0300, Oded Gabbay wrote: > > > > > > > On Tue, Sep 14, 2021 at 7:12 PM Jason Gunthorpe <jgg(a)ziepe.ca> wrote: > > > > > > > > > > > > > > > > On Tue, Sep 14, 2021 at 04:18:31PM +0200, Daniel Vetter wrote: > > > > > > > > > On Sun, Sep 12, 2021 at 07:53:07PM +0300, Oded Gabbay wrote: > > > > > > > > > > Hi, > > > > > > > > > > Re-sending this patch-set following the release of our user-space TPC > > > > > > > > > > compiler and runtime library. > > > > > > > > > > > > > > > > > > > > I would appreciate a review on this. > > > > > > > > > > > > > > > > > > I think the big open we have is the entire revoke discussions. Having the > > > > > > > > > option to let dma-buf hang around which map to random local memory ranges, > > > > > > > > > without clear ownership link and a way to kill it sounds bad to me. > > > > > > > > > > > > > > > > > > I think there's a few options: > > > > > > > > > - We require revoke support. But I've heard rdma really doesn't like that, > > > > > > > > > I guess because taking out an MR while holding the dma_resv_lock would > > > > > > > > > be an inversion, so can't be done. Jason, can you recap what exactly the > > > > > > > > > hold-up was again that makes this a no-go? > > > > > > > > > > > > > > > > RDMA HW can't do revoke. > > > > > > > > > > > > Like why? I'm assuming when the final open handle or whatever for that MR > > > > > > is closed, you do clean up everything? Or does that MR still stick around > > > > > > forever too? > > > > > > > > > > It is a combination of uAPI and HW specification. > > > > > > > > > > revoke here means you take a MR object and tell it to stop doing DMA > > > > > without causing the MR object to be destructed. > > > > > > > > > > All the drivers can of course destruct the MR, but doing such a > > > > > destruction without explicit synchronization with user space opens > > > > > things up to a serious use-after potential that could be a security > > > > > issue. > > > > > > > > > > When the open handle closes the userspace is synchronized with the > > > > > kernel and we can destruct the HW objects safely. > > > > > > > > > > So, the special HW feature required is 'stop doing DMA but keep the > > > > > object in an error state' which isn't really implemented, and doesn't > > > > > extend very well to other object types beyond simple MRs. > > > > > > > > Yeah revoke without destroying the MR doesn't work, and it sounds like > > > > revoke by destroying the MR just moves the can of worms around to another > > > > place. > > > > > > > > > > 1. User A opens gaudi device, sets up dma-buf export > > > > > > > > > > > > 2. User A registers that with RDMA, or anything else that doesn't support > > > > > > revoke. > > > > > > > > > > > > 3. User A closes gaudi device > > > > > > > > > > > > 4. User B opens gaudi device, assumes that it has full control over the > > > > > > device and uploads some secrets, which happen to end up in the dma-buf > > > > > > region user A set up > > > > > > > > > > I would expect this is blocked so long as the DMABUF exists - eg the > > > > > DMABUF will hold a fget on the FD of #1 until the DMABUF is closed, so > > > > > that #3 can't actually happen. > > > > > > > > > > > It's not mlocked memory, it's mlocked memory and I can exfiltrate > > > > > > it. > > > > > > > > > > That's just bug, don't make buggy drivers :) > > > > > > > > Well yeah, but given that habanalabs hand rolled this I can't just check > > > > for the usual things we have to enforce this in drm. And generally you can > > > > just open chardevs arbitrarily, and multiple users fighting over each > > > > another. The troubles only start when you have private state or memory > > > > allocations of some kind attached to the struct file (instead of the > > > > underlying device), or something else that requires device exclusivity. > > > > There's no standard way to do that. > > > > > > > > Plus in many cases you really want revoke on top (can't get that here > > > > unfortunately it seems), and the attempts to get towards a generic > > > > revoke() just never went anywhere. So again it's all hand-rolled > > > > per-subsystem. *insert lament about us not having done this through a > > > > proper subsystem* > > > > > > > > Anyway it sounds like the code takes care of that. > > > > -Daniel > > > > > > Daniel, Jason, > > > Thanks for reviewing this code. > > > > > > Can I get an R-B / A-B from you for this patch-set ? > > > > > > Thanks, > > > Oded > > > > A kind reminder. > > > > Thanks, > > Oded > > Hi, > I know last week was LPC and maybe this got lost in the inbox, so I'm > sending it again to make sure you got my request for R-B / A-B. I was waiting for some clarity from the maintainers summit, but that's still about as unclear as it gets. Either way technically it sounds ok, but I'm a bit burried so didn't look at the code. Acked-by: Daniel Vetter <daniel.vetter(a)ffwll.ch> But looking beyond the strict lens of dma-buf I'm still impressed by the mess this created, to get to the same endpoint of "we open our stack" in the same time it takes others to sort this out. I'm still looking for some kind of plan to fix this. Also you probably want to get Dave to ack this too, I pinged him on irc last week about this after maintainer summit. -Daniel -- Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch

2 years, 6 months

Re: [Linaro-mm-sig] [PATCH v6 2/2] habanalabs: add support for dma-buf exporter

by Jason Gunthorpe

On Sun, Sep 12, 2021 at 07:53:09PM +0300, Oded Gabbay wrote: > From: Tomer Tayar <ttayar(a)habana.ai> > > Implement the calls to the dma-buf kernel api to create a dma-buf > object backed by FD. > > We block the option to mmap the DMA-BUF object because we don't support > DIRECT_IO and implicit P2P. This statement doesn't make sense, you can mmap your dmabuf if you like. All dmabuf mmaps are supposed to set the special bit/etc to exclude them from get_user_pages() anyhow - and since this is BAR memory not struct page memory this driver would be doing it anyhow. > We check the p2p distance using pci_p2pdma_distance_many() and refusing > to map dmabuf in case the distance doesn't allow p2p. Does this actually allow the p2p transfer for your intended use cases? > diff --git a/drivers/misc/habanalabs/common/memory.c b/drivers/misc/habanalabs/common/memory.c > index 33986933aa9e..8cf5437c0390 100644 > +++ b/drivers/misc/habanalabs/common/memory.c > @@ -1,7 +1,7 @@ > // SPDX-License-Identifier: GPL-2.0 > > /* > - * Copyright 2016-2019 HabanaLabs, Ltd. > + * Copyright 2016-2021 HabanaLabs, Ltd. > * All Rights Reserved. > */ > > @@ -11,11 +11,13 @@ > > #include <linux/uaccess.h> > #include <linux/slab.h> > +#include <linux/pci-p2pdma.h> > > #define HL_MMU_DEBUG 0 > > /* use small pages for supporting non-pow2 (32M/40M/48M) DRAM phys page sizes */ > -#define DRAM_POOL_PAGE_SIZE SZ_8M > +#define DRAM_POOL_PAGE_SIZE SZ_8M > + ?? > /* > * The va ranges in context object contain a list with the available chunks of > @@ -347,6 +349,13 @@ static int free_device_memory(struct hl_ctx *ctx, struct hl_mem_in *args) > return -EINVAL; > } > > + if (phys_pg_pack->exporting_cnt) { > + dev_err(hdev->dev, > + "handle %u is exported, cannot free\n", handle); > + spin_unlock(&vm->idr_lock); Don't write to the kernel log from user space triggered actions > +static int alloc_sgt_from_device_pages(struct hl_device *hdev, > + struct sg_table **sgt, u64 *pages, > + u64 npages, u64 page_size, > + struct device *dev, > + enum dma_data_direction dir) Why doesn't this return a sg_table * and an ERR_PTR? > +{ > + u64 chunk_size, bar_address, dma_max_seg_size; > + struct asic_fixed_properties *prop; > + int rc, i, j, nents, cur_page; > + struct scatterlist *sg; > + > + prop = &hdev->asic_prop; > + > + dma_max_seg_size = dma_get_max_seg_size(dev); > + > + /* We would like to align the max segment size to PAGE_SIZE, so the > + * SGL will contain aligned addresses that can be easily mapped to > + * an MMU > + */ > + dma_max_seg_size = ALIGN_DOWN(dma_max_seg_size, PAGE_SIZE); > + if (dma_max_seg_size < PAGE_SIZE) { > + dev_err_ratelimited(hdev->dev, > + "dma_max_seg_size %llu can't be smaller than PAGE_SIZE\n", > + dma_max_seg_size); > + return -EINVAL; > + } > + > + *sgt = kzalloc(sizeof(**sgt), GFP_KERNEL); > + if (!*sgt) > + return -ENOMEM; > + > + /* If the size of each page is larger than the dma max segment size, > + * then we can't combine pages and the number of entries in the SGL > + * will just be the > + * <number of pages> * <chunks of max segment size in each page> > + */ > + if (page_size > dma_max_seg_size) > + nents = npages * DIV_ROUND_UP_ULL(page_size, dma_max_seg_size); > + else > + /* Get number of non-contiguous chunks */ > + for (i = 1, nents = 1, chunk_size = page_size ; i < npages ; i++) { > + if (pages[i - 1] + page_size != pages[i] || > + chunk_size + page_size > dma_max_seg_size) { > + nents++; > + chunk_size = page_size; > + continue; > + } > + > + chunk_size += page_size; > + } > + > + rc = sg_alloc_table(*sgt, nents, GFP_KERNEL | __GFP_ZERO); > + if (rc) > + goto error_free; > + > + /* Because we are not going to include a CPU list we want to have some > + * chance that other users will detect this by setting the orig_nents > + * to 0 and using only nents (length of DMA list) when going over the > + * sgl > + */ > + (*sgt)->orig_nents = 0; Maybe do this at the end so you'd have to undo it on the error path? > + cur_page = 0; > + > + if (page_size > dma_max_seg_size) { > + u64 size_left, cur_device_address = 0; > + > + size_left = page_size; > + > + /* Need to split each page into the number of chunks of > + * dma_max_seg_size > + */ > + for_each_sgtable_dma_sg((*sgt), sg, i) { > + if (size_left == page_size) > + cur_device_address = > + pages[cur_page] - prop->dram_base_address; > + else > + cur_device_address += dma_max_seg_size; > + > + chunk_size = min(size_left, dma_max_seg_size); > + > + bar_address = hdev->dram_pci_bar_start + cur_device_address; > + > + rc = set_dma_sg(sg, bar_address, chunk_size, dev, dir); > + if (rc) > + goto error_unmap; > + > + if (size_left > dma_max_seg_size) { > + size_left -= dma_max_seg_size; > + } else { > + cur_page++; > + size_left = page_size; > + } > + } > + } else { > + /* Merge pages and put them into the scatterlist */ > + for_each_sgtable_dma_sg((*sgt), sg, i) { > + chunk_size = page_size; > + for (j = cur_page + 1 ; j < npages ; j++) { > + if (pages[j - 1] + page_size != pages[j] || > + chunk_size + page_size > dma_max_seg_size) > + break; > + > + chunk_size += page_size; > + } > + > + bar_address = hdev->dram_pci_bar_start + > + (pages[cur_page] - prop->dram_base_address); > + > + rc = set_dma_sg(sg, bar_address, chunk_size, dev, dir); > + if (rc) > + goto error_unmap; > + > + cur_page = j; > + } > + } We have this sg_append stuff now that is intended to help building these things. It can only build CPU page lists, not these DMA lists, but I do wonder if open coding in drivers is slipping back a bit. Especially since AMD seems to be doing something different. Could the DMABUF layer gain some helpers styled after the sg_append to simplify building these things? and convert the AMD driver of course. > +static int hl_dmabuf_attach(struct dma_buf *dmabuf, > + struct dma_buf_attachment *attachment) > +{ > + struct hl_dmabuf_wrapper *hl_dmabuf; > + struct hl_device *hdev; > + int rc; > + > + hl_dmabuf = dmabuf->priv; > + hdev = hl_dmabuf->ctx->hdev; > + > + rc = pci_p2pdma_distance_many(hdev->pdev, &attachment->dev, 1, true); > + > + if (rc < 0) > + attachment->peer2peer = false; Extra blank line > + > + return 0; > +} > + > +static struct sg_table *hl_map_dmabuf(struct dma_buf_attachment *attachment, > + enum dma_data_direction dir) > +{ > + struct dma_buf *dma_buf = attachment->dmabuf; > + struct hl_vm_phys_pg_pack *phys_pg_pack; > + struct hl_dmabuf_wrapper *hl_dmabuf; > + struct hl_device *hdev; > + struct sg_table *sgt; > + int rc; > + > + hl_dmabuf = dma_buf->priv; > + hdev = hl_dmabuf->ctx->hdev; > + phys_pg_pack = hl_dmabuf->phys_pg_pack; > + > + if (!attachment->peer2peer) { > + dev_err(hdev->dev, > + "Failed to map dmabuf because p2p is disabled\n"); > + return ERR_PTR(-EPERM); User triggered printable again? > +static void hl_unmap_dmabuf(struct dma_buf_attachment *attachment, > + struct sg_table *sgt, > + enum dma_data_direction dir) > +{ > + struct scatterlist *sg; > + int i; > + > + for_each_sgtable_dma_sg(sgt, sg, i) > + dma_unmap_resource(attachment->dev, sg_dma_address(sg), > + sg_dma_len(sg), dir, > + DMA_ATTR_SKIP_CPU_SYNC); Why can we skip the CPU_SYNC? Seems like a comment is needed Something has to do a CPU_SYNC before recylcing this memory for another purpose, where is it? > +static void hl_release_dmabuf(struct dma_buf *dmabuf) > +{ > + struct hl_dmabuf_wrapper *hl_dmabuf = dmabuf->priv; Maybe hl_dmabuf_wrapper should be hl_dmabuf_priv > + * export_dmabuf_from_addr() - export a dma-buf object for the given memory > + * address and size. > + * @ctx: pointer to the context structure. > + * @device_addr: device memory physical address. > + * @size: size of device memory. > + * @flags: DMA-BUF file/FD flags. > + * @dmabuf_fd: pointer to result FD that represents the dma-buf object. > + * > + * Create and export a dma-buf object for an existing memory allocation inside > + * the device memory, and return a FD which is associated with the dma-buf > + * object. > + * > + * Return: 0 on success, non-zero for failure. > + */ > +static int export_dmabuf_from_addr(struct hl_ctx *ctx, u64 device_addr, > + u64 size, int flags, int *dmabuf_fd) > +{ > + struct hl_dmabuf_wrapper *hl_dmabuf; > + struct hl_device *hdev = ctx->hdev; > + struct asic_fixed_properties *prop; > + u64 bar_address; > + int rc; > + > + prop = &hdev->asic_prop; > + > + if (!IS_ALIGNED(device_addr, PAGE_SIZE)) { > + dev_err_ratelimited(hdev->dev, > + "address of exported device memory should be aligned to 0x%lx, address 0x%llx\n", > + PAGE_SIZE, device_addr); > + return -EINVAL; > + } > + > + if (size < PAGE_SIZE) { > + dev_err_ratelimited(hdev->dev, > + "size %llu of exported device memory should be equal to or greater than %lu\n", > + size, PAGE_SIZE); > + return -EINVAL; > + } > + > + if (device_addr < prop->dram_user_base_address || > + device_addr + size > prop->dram_end_address || > + device_addr + size < device_addr) { > + dev_err_ratelimited(hdev->dev, > + "DRAM memory range is outside of DRAM boundaries, address 0x%llx, size 0x%llx\n", > + device_addr, size); > + return -EINVAL; > + } > + > + bar_address = hdev->dram_pci_bar_start + > + (device_addr - prop->dram_base_address); > + > + if (bar_address + size > > + hdev->dram_pci_bar_start + prop->dram_pci_bar_size || > + bar_address + size < bar_address) { > + dev_err_ratelimited(hdev->dev, > + "DRAM memory range is outside of PCI BAR boundaries, address 0x%llx, size 0x%llx\n", > + device_addr, size); > + return -EINVAL; > + } More prints from userspace > +static int export_dmabuf_from_handle(struct hl_ctx *ctx, u64 handle, int flags, > + int *dmabuf_fd) > +{ > + struct hl_vm_phys_pg_pack *phys_pg_pack; > + struct hl_dmabuf_wrapper *hl_dmabuf; > + struct hl_device *hdev = ctx->hdev; > + struct asic_fixed_properties *prop; > + struct hl_vm *vm = &hdev->vm; > + u64 bar_address; > + u32 idr_handle; > + int rc, i; > + > + prop = &hdev->asic_prop; > + > + idr_handle = lower_32_bits(handle); Why silent truncation? Shouldn't setting the upper 32 bits be an error? > + case HL_MEM_OP_EXPORT_DMABUF_FD: > + rc = export_dmabuf_from_addr(ctx, > + args->in.export_dmabuf_fd.handle, > + args->in.export_dmabuf_fd.mem_size, > + args->in.flags, > + &dmabuf_fd); > + memset(args, 0, sizeof(*args)); > + args->out.fd = dmabuf_fd; Would expect the installed fd to be the positive return, not a pointer Jason

2 years, 7 months

Re: [Linaro-mm-sig] [PATCH v6 1/2] habanalabs: define uAPI to export FD for DMA-BUF

by Jason Gunthorpe

On Sun, Sep 12, 2021 at 07:53:08PM +0300, Oded Gabbay wrote: > /* HL_MEM_OP_* */ > __u32 op; > - /* HL_MEM_* flags */ > + /* HL_MEM_* flags. > + * For the HL_MEM_OP_EXPORT_DMABUF_FD opcode, this field holds the > + * DMA-BUF file/FD flags. > + */ > __u32 flags; > /* Context ID - Currently not in use */ > __u32 ctx_id; > @@ -1072,6 +1091,13 @@ struct hl_mem_out { > > __u32 pad; > }; > + > + /* Returned in HL_MEM_OP_EXPORT_DMABUF_FD. Represents the > + * DMA-BUF object that was created to describe a memory > + * allocation on the device's memory space. The FD should be > + * passed to the importer driver > + */ > + __u64 fd; fd's should be a s32 type in a fixed width uapi. I usually expect to see the uapi changes inside the commit that consumes them, splitting the patch like this seems strange but harmless. Jason

2 years, 7 months

[PATCH v3 0/9] dma-fence: Deadline awareness

by Rob Clark

From: Rob Clark <robdclark(a)chromium.org> This series adds deadline awareness to fences, so realtime deadlines such as vblank can be communicated to the fence signaller for power/ frequency management decisions. This is partially inspired by a trick i915 does, but implemented via dma-fence for a couple of reasons: 1) To continue to be able to use the atomic helpers 2) To support cases where display and gpu are different drivers This iteration adds a dma-fence ioctl to set a deadline (both to support igt-tests, and compositors which delay decisions about which client buffer to display), and a sw_sync ioctl to read back the deadline. IGT tests utilizing these can be found at: https://gitlab.freedesktop.org/robclark/igt-gpu-tools/-/commits/fence-deadl… v1: https://patchwork.freedesktop.org/series/93035/ v2: Move filtering out of later deadlines to fence implementation to avoid increasing the size of dma_fence v3: Add support in fence-array and fence-chain; Add some uabi to support igt tests and userspace compositors. Rob Clark (9): dma-fence: Add deadline awareness drm/vblank: Add helper to get next vblank time drm/atomic-helper: Set fence deadline for vblank drm/scheduler: Add fence deadline support drm/msm: Add deadline based boost support dma-buf/fence-array: Add fence deadline support dma-buf/fence-chain: Add fence deadline support dma-buf/sync_file: Add SET_DEADLINE ioctl dma-buf/sw_sync: Add fence deadline support drivers/dma-buf/dma-fence-array.c | 11 ++++ drivers/dma-buf/dma-fence-chain.c | 13 +++++ drivers/dma-buf/dma-fence.c | 20 +++++++ drivers/dma-buf/sw_sync.c | 58 +++++++++++++++++++ drivers/dma-buf/sync_debug.h | 2 + drivers/dma-buf/sync_file.c | 19 +++++++ drivers/gpu/drm/drm_atomic_helper.c | 36 ++++++++++++ drivers/gpu/drm/drm_vblank.c | 32 +++++++++++ drivers/gpu/drm/msm/msm_fence.c | 76 +++++++++++++++++++++++++ drivers/gpu/drm/msm/msm_fence.h | 20 +++++++ drivers/gpu/drm/msm/msm_gpu.h | 1 + drivers/gpu/drm/msm/msm_gpu_devfreq.c | 20 +++++++ drivers/gpu/drm/scheduler/sched_fence.c | 34 +++++++++++ drivers/gpu/drm/scheduler/sched_main.c | 2 +- include/drm/drm_vblank.h | 1 + include/drm/gpu_scheduler.h | 8 +++ include/linux/dma-fence.h | 16 ++++++ include/uapi/linux/sync_file.h | 20 +++++++ 18 files changed, 388 insertions(+), 1 deletion(-) -- 2.31.1

2 years, 7 months

[PATCH 1/4] dma-buf: add dma_fence_describe and dma_resv_describe

by Christian König

Add functions to dump dma_fence and dma_resv objects into a seq_file and use them for printing the debugfs informations. Signed-off-by: Christian König <christian.koenig(a)amd.com> --- drivers/dma-buf/dma-buf.c | 11 +---------- drivers/dma-buf/dma-fence.c | 16 ++++++++++++++++ drivers/dma-buf/dma-resv.c | 23 +++++++++++++++++++++++ include/linux/dma-fence.h | 1 + include/linux/dma-resv.h | 1 + 5 files changed, 42 insertions(+), 10 deletions(-) diff --git a/drivers/dma-buf/dma-buf.c b/drivers/dma-buf/dma-buf.c index d35c71743ccb..4975c9289b02 100644 --- a/drivers/dma-buf/dma-buf.c +++ b/drivers/dma-buf/dma-buf.c @@ -1368,8 +1368,6 @@ static int dma_buf_debug_show(struct seq_file *s, void *unused) { struct dma_buf *buf_obj; struct dma_buf_attachment *attach_obj; - struct dma_resv_iter cursor; - struct dma_fence *fence; int count = 0, attach_count; size_t size = 0; int ret; @@ -1397,14 +1395,7 @@ static int dma_buf_debug_show(struct seq_file *s, void *unused) file_inode(buf_obj->file)->i_ino, buf_obj->name ?: ""); - dma_resv_for_each_fence(&cursor, buf_obj->resv, true, fence) { - seq_printf(s, "\t%s fence: %s %s %ssignalled\n", - dma_resv_iter_is_exclusive(&cursor) ? - "Exclusive" : "Shared", - fence->ops->get_driver_name(fence), - fence->ops->get_timeline_name(fence), - dma_fence_is_signaled(fence) ? "" : "un"); - } + dma_resv_describe(buf_obj->resv, s); seq_puts(s, "\tAttached Devices:\n"); attach_count = 0; diff --git a/drivers/dma-buf/dma-fence.c b/drivers/dma-buf/dma-fence.c index 1e82ecd443fa..5175adf58644 100644 --- a/drivers/dma-buf/dma-fence.c +++ b/drivers/dma-buf/dma-fence.c @@ -907,6 +907,22 @@ dma_fence_wait_any_timeout(struct dma_fence **fences, uint32_t count, } EXPORT_SYMBOL(dma_fence_wait_any_timeout); +/** + * dma_fence_describe - Dump fence describtion into seq_file + * @fence: the 6fence to describe + * @seq: the seq_file to put the textual description into + * + * Dump a textual description of the fence and it's state into the seq_file. + */ +void dma_fence_describe(struct dma_fence *fence, struct seq_file *seq) +{ + seq_printf(seq, "%s %s seq %llu %ssignalled\n", + fence->ops->get_driver_name(fence), + fence->ops->get_timeline_name(fence), fence->seqno, + dma_fence_is_signaled(fence) ? "" : "un"); +} +EXPORT_SYMBOL(dma_fence_describe); + /** * dma_fence_init - Initialize a custom fence. * @fence: the fence to initialize diff --git a/drivers/dma-buf/dma-resv.c b/drivers/dma-buf/dma-resv.c index 266ec9e3caef..6bb25d53e702 100644 --- a/drivers/dma-buf/dma-resv.c +++ b/drivers/dma-buf/dma-resv.c @@ -38,6 +38,7 @@ #include <linux/mm.h> #include <linux/sched/mm.h> #include <linux/mmu_notifier.h> +#include <linux/seq_file.h> /** * DOC: Reservation Object Overview @@ -654,6 +655,28 @@ bool dma_resv_test_signaled(struct dma_resv *obj, bool test_all) } EXPORT_SYMBOL_GPL(dma_resv_test_signaled); +/** + * dma_resv_describe - Dump description of the resv object into seq_file + * @obj: the reservation object + * @seq: the seq_file to dump the description into + * + * Dump a textual description of the fences inside an dma_resv object into the + * seq_file. + */ +void dma_resv_describe(struct dma_resv *obj, struct seq_file *seq) +{ + struct dma_resv_iter cursor; + struct dma_fence *fence; + + dma_resv_for_each_fence(&cursor, obj, true, fence) { + seq_printf(seq, "\t%s fence:", + dma_resv_iter_is_exclusive(&cursor) ? + "Exclusive" : "Shared"); + dma_fence_describe(fence, seq); + } +} +EXPORT_SYMBOL_GPL(dma_resv_describe); + #if IS_ENABLED(CONFIG_LOCKDEP) static int __init dma_resv_lockdep(void) { diff --git a/include/linux/dma-fence.h b/include/linux/dma-fence.h index a706b7bf51d7..1ea691753bd3 100644 --- a/include/linux/dma-fence.h +++ b/include/linux/dma-fence.h @@ -264,6 +264,7 @@ void dma_fence_init(struct dma_fence *fence, const struct dma_fence_ops *ops, void dma_fence_release(struct kref *kref); void dma_fence_free(struct dma_fence *fence); +void dma_fence_describe(struct dma_fence *fence, struct seq_file *seq); /** * dma_fence_put - decreases refcount of the fence diff --git a/include/linux/dma-resv.h b/include/linux/dma-resv.h index d4b4cd43f0f1..49c0152073fd 100644 --- a/include/linux/dma-resv.h +++ b/include/linux/dma-resv.h @@ -486,5 +486,6 @@ int dma_resv_copy_fences(struct dma_resv *dst, struct dma_resv *src); long dma_resv_wait_timeout(struct dma_resv *obj, bool wait_all, bool intr, unsigned long timeout); bool dma_resv_test_signaled(struct dma_resv *obj, bool test_all); +void dma_resv_describe(struct dma_resv *obj, struct seq_file *seq); #endif /* _LINUX_RESERVATION_H */ -- 2.25.1

2 years, 7 months

Re: [Linaro-mm-sig] [PATCH net-next 3/5] net: bcmasp: Add support for ASP2.0 Ethernet controller

by Andrew Lunn

> > > +static int bcmasp_set_priv_flags(struct net_device *dev, u32 flags) > > > +{ > > > + struct bcmasp_intf *intf = netdev_priv(dev); > > > + > > > + intf->wol_keep_rx_en = flags & BCMASP_WOL_KEEP_RX_EN ? 1 : 0; > > > + > > > + return 0; > > > > Please could you explain this some more. How can you disable RX and > > still have WoL working? > > Wake-on-LAN using Magic Packets and network filters requires keeping the > UniMAC's receiver turned on, and then the packets feed into the Magic Packet > Detector (MPD) block or the network filter block. In that mode DRAM is in > self refresh and there is local matching of frames into a tiny FIFO however > in the case of magic packets the packets leading to a wake-up are dropped as > there is nowhere to store them. In the case of a network filter match (e.g.: > matching a multicast IP address plus protocol, plus source/destination > ports) the packets are also discarded because the receive DMA was shut down. > > When the wol_keep_rx_en flag is set, the above happens but we also allow the > packets that did match a network filter to reach the small FIFO (Justin > would know how many entries are there) that is used to push the packets to > DRAM. The packet contents are held in there until the system wakes up which > is usually just a few hundreds of micro seconds after we received a packet > that triggered a wake-up. Once we overflow the receive DMA FIFO capacity > subsequent packets get dropped which is fine since we are usually talking > about very low bit rates, and we only try to push to DRAM the packets of > interest, that is those for which we have a network filter. > > This is convenient in scenarios where you want to wake-up from multicast DNS > (e.g.: wake on Googlecast, Bonjour etc.) because then the packet that > resulted in the system wake-up is not discarded but is then delivered to the > network stack. Thanks for the explanation. It would be easier for the user if you automate this. Enable is by default for WoL types which have user content? > > > + /* Per ch */ > > > + intf->tx_spb_dma = priv->base + TX_SPB_DMA_OFFSET(intf); > > > + intf->res.tx_spb_ctrl = priv->base + TX_SPB_CTRL_OFFSET(intf); > > > + /* > > > + * Stop gap solution. This should be removed when 72165a0 is > > > + * deprecated > > > + */ > > > > Is that an internal commit? > > Yes this is a revision of the silicon that is not meant to see the light of > day. So this can all be removed? Andrew

2 years, 7 months

Re: [Linaro-mm-sig] [PATCH net-next 2/5] dt-bindings: net: brcm, unimac-mdio: Add asp-v2.0

by Rob Herring

On Fri, 24 Sep 2021 14:44:48 -0700, Justin Chen wrote: > The ASP 2.0 Ethernet controller uses a brcm unimac. > > Signed-off-by: Justin Chen <justinpopo6(a)gmail.com> > Signed-off-by: Florian Fainelli <f.fainelli(a)gmail.com> > --- > Documentation/devicetree/bindings/net/brcm,unimac-mdio.yaml | 1 + > 1 file changed, 1 insertion(+) > Running 'make dtbs_check' with the schema in this patch gives the following warnings. Consider if they are expected or the schema is incorrect. These may not be new warnings. Note that it is not yet a requirement to have 0 warnings for dtbs_check. This will change in the future. Full log is available here: https://patchwork.ozlabs.org/patch/1532529 mdio@e14: #address-cells:0:0: 1 was expected arch/arm64/boot/dts/broadcom/bcm2711-rpi-400.dt.yaml arch/arm64/boot/dts/broadcom/bcm2711-rpi-4-b.dt.yaml arch/arm/boot/dts/bcm2711-rpi-400.dt.yaml arch/arm/boot/dts/bcm2711-rpi-4-b.dt.yaml mdio@e14: #size-cells:0:0: 0 was expected arch/arm64/boot/dts/broadcom/bcm2711-rpi-400.dt.yaml arch/arm64/boot/dts/broadcom/bcm2711-rpi-4-b.dt.yaml arch/arm/boot/dts/bcm2711-rpi-400.dt.yaml arch/arm/boot/dts/bcm2711-rpi-4-b.dt.yaml

2 years, 7 months

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

Linaro-mm-sig September 2021